Building Features and Machine Learning Models for Audio Signals
summary
Python library pi_audio_processing provides MFCC and GFCC extraction plus spectral and chroma features, end-to-end cleaning (low-activity removal, spectral subtraction) and scikit-learn classifiers (SVM, RF, LR, KNN, GB). Experiments use 150 samples (50 each) for speech/music/birds with test results like 13/14 music and near-perfect speech, GTZAN genre (80/20 per class) improves by nine correct labels when adding spectral+chroma to mfcc, and a location task climbs from 8/17 (female-only train, male test) to 15/17 when combining gfcc and spectral features.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:25Agenda
- 2:52Machine Learning Overview
- 4:54What is Audio
- 6:59Audio Transformation
- 14:31Tools
- 15:18Audio Processing
- 16:33Low Activity Segment Removal
- 17:20Data Collection
- 21:50Audio Type Classification
- 23:44Music Genre Classification
- 27:32Location Classification
- 29:49QA
related talks
31mClassifying Audio into Types using Python
11mClassifying Spoken Location Names
40mGathering Insights from Audio Data
29mRecognizing Speech in Python
25mLooking at Sound
51mThe Merger of Machine Learning, Music and Programming
22mImplementation and analysis of deep learning models for speech classification
40mWho's singing? Automatic bird sound recognition with machine learning